If you are reading this and happen to be an eSports player, all I can say is ...... that you've been running a lot. Because in one legendary driving simulation series, the technology has advanced to the point where the AI can hand over the kester to a professional gamer. [Enter GT Sophy. It is a form of machine learning that teaches the neural network of an AI system through rewards and reprimands (thanks to "New Scientist").

Using both carrots and sticks, GT Sophy managed to beat 95% of the players on the online leaderboards of Gran Turismo Sport after just a few days.

Of course, the AI is trained to have advantages that human players do not have; GT Sophy knows the location of all rival cars in the PS4 driving simulator during a race, and at any given moment also has data on the course for the next six seconds. On the flip side, Sony only teaches GT Sophy to auto shift, whereas the top players in Gran Turismo can shift manually.

Despite the innate advantage, it took 45,000,000 hours of virtual driving to get GT Sophy to drive a particular course faster than any human on the planet. Also, GT Sophy plays on 20 PS4s at a time.

Another major pain encountered by the AI was learning how to outrun rival racers. Sony's team solved this by rewarding GT Sophy when it overtook cars and penalizing it when it was overtaken.

Sony's AI researchers also began teaming up with competitive Gran Turismo players to focus on teaching GT Sophy how to master the trickiest sections from the game's courses.

At this point, the AI was ready to take on world-class players. So the researchers pitted four Sophie's against four of the best GT drivers in a series of races. The result? GT Sophie beat the top human racers by a margin of 104 to 54.

The moral of the story, thankfully, does not end with "long live Skynet." According to Igor Babuskin of OpenAI in San Francisco, the deep reinforcement learning method that trained GT Sophy could be used by future developers to give in-game teammates and NPCs more intelligent AI for players to interact with." The results suggest that game developers could use deep reinforcement learning to design and test games to create interesting opponents and teammates for human players."[15

On the other hand, if I were a Gran Turismo 7 player, I would spend at least a couple of hours every day practicing my turns at Deep Forest Raceway before GT Sophy steps up to shame the entire community of PS5 racers.